Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
~6 measured is in fact endogenous. In addition to these problems of com- parability and causal placement, the WFS measure of desired family size has questionable reliability (Westoff, 1980:2) and an unknown correspon- dence to underlying preferences (Szykman, 1982~. Our model also omits the costs of fertility control. Easterlin (1978) defines three types of cost: fixed costs, which might be tapped by knowledge of contraception; variable costs, which refer to the recurring expense of purchasing contraceptive supplies; and psychic costs, which involve attitudes toward and norms governing contraceptive use. m e higher these costs, the less likely contraceptive use will be. Although the WFS standard recode tapes include indicators of contracep- tive knowledge, they lack information about variable and psychic costs. It is doubtful that adding either desired family size or cost factors would much improve the predictive power of the model. If appropriate and reliable indicators were available, we would place these variables in the causal structure so that they mediated the influence of socioeconomic variables on later fertility and contraceptive use. Weir effects are, in our model, implicit in those of the socioeconomic variables. The model could easily be modified to accommodate these variables should better information become available. Finally, given that the WFS countries are located along the traditional/transitional continuum, it is plausible to view desired family size and fertility regulation costs as relatively invariant either within countries or for well-defined social groups within countries. If so, it would be conceptually appropriate to account for these two vari- ables at the macro level. As noted earlier, we plan to construct aggre- gated or global macro variables that can be used to analyze varability across settings in the micro-level coefficients of the model. 1.3 THE STRUCTURAL EQUATIONS This section explains the sixteen structural equations of the processual model, as well as our expectations about the nature and direction of the relationships involved. Because these expectations depend on societal context, we continue to invoke the ideal-typic traditional/transitional distinction described above. Table 1.2 summarizes these cross-setting expectations about the signs of all coefficients in the structural equations. The following discussion also provides relatively precise definitions of the variables comprising the model, all of which depend on data available from WFS standard recode tapes. For simplicity, each structural equation is treated as if it were estimable as a multiple regression. Although actual estimation might be considerably more complex, allowance here for details specific to discrete and ordinal response variables would detract from explication of the substantive hypotheses underlying the model.3 Onset Taken by themselves, Blocks I and II of Figure 1.2 depict a model of fertility onset, defined as age at first birth (AFB). His provides an
17 _ a. · - In U) o I) U] MU o U] o . - U) Ed o · - En In . - - ha A: s En · - P. . En . 3 un _ en Cal a _ 8- _ o~ ~ 0 0 ~ :r: _ ~ _ :C £ _ U] 0 3 _ c: U) m_ In - ~ x: ~ _ m ~ A: - u, Q 1 + 1 1 1 1 1 1 1 + 1 + + 0 0 + O 0 + 0 + + O 0 0 0 0 0 + + + + 1 + + + + + + + 1 + + 1 + + cI' ~ o o o c) ~ o c' o c' ~ ~ o ~ o o o o o o o I I + + + ic ~ t I I + I + o o + + + + + 1 0 + + + 1 0 + + + I + I ~ C) ~ C) ~ ~ O V ~ ~ ~ ~ m ~ ~n £ ~ a: ~ ~ * 4: ~ a' R3 ~ ~ ~ ~ O t) 93 u~ ~ ~ ~ 3 tt: 3 ~: 3 W ~ u~ u~ P~ Z a ~n3 ~ u~ u · - JJ ~ · · - ) ~ ~ ~ C · - - n ~ C ~ a' ~ ~ u' · - £ - s ~5 C O O ~ O - ~ C ~Q 0= ~ C O ~n C £ C Y O U] C ~ O · ~ 54 ·,~ U] · - C eq O · ~ ~ :: ~: . - c: u] ~ · - .-' ~ · ~ c: ~ u' ~ o u) ~ ~ ~ c ~ ~ o ac ~ 0 `~ - ;= 0 ~ ~ ~ ·~~ ~~ ~ c: o o ~ ~ c: ~ . - ~ ~ c) .~' ~ · - ~n .-l ~ ~ ~ ~n ~ c `: ~ ~ ~ o . - · - U .-l ~ . - . - o ] - `: ~ ~ o ~ ' eq ·c =~ - x ~ ~ cn 0 ~ ~ o ~ .e s C: JJ · ~= 0 'Q C: eq t) O ~ ~ N eQ ~ O ~ ~ O o 3 ac Q ~ 0 O ~n 0 ·,~ .~: 0 u, 0 u, ~ x ~: ~ ~ u' s ~ ~ ~c .. ~ ~ O ~ ~ ~ O ~ O ~ ~ ~ oo 3 ~ ~ ~
18 unambiguous starting point for the fertility process that is comparable across cultures. Since contraceptive use between entry into a union and the birth of the first child is not expected in WFS countries, AFB is a simple function of age at first union and a delay influenced by fecund- ability and chance factors. Consequently, hypotheses about the socio- economic determinants of age at first union can be applied directly to AFB. Age at first union corresponds to age at marriage in many, although not all, WFS countries. The onset model contains a second endogenous variable: respondent work experience prior to marriage (WBM)e The focus here is on whether respondents work away from home in a nonfamily enterprise in order to measure participation in the modern sector and exposure to modern norms and ideas. WBM is dichotomous and distinguishes between experience in the modern sector and all other work possibilities. m is reflects a constraint of the WFS data. Although it would be helpful to know how long respondents were employed before marriage and how modern that employment was, this information is unavailable in the WFS data, as well as in most other fertility surveys. Figure 1.2 displays AFB and WBM as contemporaneously correlated variables in Block II. This is a deliberate departure from the seemingly obvious causal ordering in which AFB depends on WBM. Formal marriage can occur after the first birth, thus disrupting the implied temporal sequence of work, marriage, and first birth. More importantly, the discretional component of AFB is the decision to enter a union, and WBM is not predetermined with respect to that decision. Not only does work before marriage imply later entry into a union, but delayed entry may also necessitate or provide the opportunity for work. This simultaneity underlies the contemporaneous placement of WBM and AFB. However, we will not estimate a simultaneous equations model because instrumental variables are not available (i.e., we would be unable to identify the coefficients for these two equations). Moreover, our purpose is to study the role of onset in mediating the effects of socioeconomic variables on family size, child mortality, and contraceptive use, rather than to address the simul- taneities within the onset model per se. Thus there would be no real loss in omitting this particular simultaneity even if it were estimable. Block I contains the only socioeconomic and sociodemographic vari- ables available for all WFS data sets which are unambiguously predetero mined with respect to AFB and WBM: respondent education (WED) and respondent childhood residence (RESC). Together, Blocks ~ and II imply the following equations: (1.1) AFB = 60,1 + 51,tRE~ + 82,1WED- (1.2) WBM = 60,2 + 61,2RESC + 02~2 For each coefficient' ~ i, ·. the i-subscript denotes the variable number and the -subscript the equation number. For most WFS data sets, we will treat respondent childhood residence as a rural-urban dichotomy (urban = 1; rural = 0), with additional detail incorporated when available (e.g., a trichotomy distinguishing countryside, town, and urban childhood residence). For ease of exposition, we will assume that respondent
19 education is scaled in years of schooling, although the model does not require this treatment, and our empirical work will explore alternatives. It is widely accepted that education is a measure of socioeconomic status (SES). In the present circumstances, it is plausible to treat place of residence similarly. The WFS data, which are notable for their lack of exogenous socioeconomic variables, do not contain information on parental SES. Urban-rural variability in childhood residence can approxi mate between-fam~ly differences in childhood SES to a degree. Moreover, in transitional settings, this variability in childhood residence is likely to reflect differential experiences In educational quality, expo- sure to relatively industrialized labor markets, and, in general, degree of exposure to relatively wealthier life styles. These and other points will be raised below in the discussion of the potential effects of child- hood residence on AFB and WBM. Both childhood residence and education should affect AFB, the nature of the hypothesized effects depending on context. In -traditional contexts, relatively advantaged women (urban child- hood residence, more education) should have a lower mean age at marriage and at first birth. Education and childhood residence reflect wealth and knowledge, which in turn affect nutritional levels, health-related behavior, and exposure to disease. Frisch (1975) and others have argued that nutrition and health are positively related to fecundability and inversely related to conception delay (although there is debate on this point; see Bongaarts, 1980~. We hypothesize that in traditional contexts, the health consequences of education and childhood residence are the critical factors in the interval between union and first birth. Education and childhood residence cannot reflect participation in modern labor markets since these are largely absent in the ideal-typic traditional setting. The hypotheses about the effects of childhood residence and education on AFB in transitional contexts are quite different. Relatively advan- taged women may still have higher monthly conception probabilities than their less advantaged counterparts. The effects of health and nutrition may, however, be overshadowed by education and childhood residence differ- entials in age at marriage. Durch (1980:18-19) shows that education and urban childhood residence vary positively with mean age at marriage in almost all of the 15 WFS countries under consideration. Childhood residence and education co-vary with the distribution of norms governing marriage patterns. According to Caldwell (1976), education represents a major conduit for the transmission of Western ideals and norms (including those that affect the timing of entry into unions) to women in less- developed countries. Childhood residence would also affect exposure to such norms and ideals. The length of time needed to find an appropriate husband may also vary positively with education. Bowever, since it is unlikely that education competes directly with marriage or motherhood in WFS countries, education is justifiably a predetermined variable with respect to AFB. In sum, in transitional settings, we expect positive relationships between WED and AFB and between RESC and AFB. Finally, we hypothesize that in transitional societies, childhood residence and education will have positive effects on the probability of working in the modern sector before marriage. Modern employment requires
20 that several conditions be met. The first is awareness. Respondents may learn of modern employment opportunities in school, or, depending on childhood residence, may have observed the participation of other com- munity members. Second, respondents must qualify for modern employment, a condition in which education plays a major role. Third, respondents must be willing to accept such jobs when offered. Norm" regarding the desirability of extrafamilial employment may be communicated in the classroom. Norms prevailing during childhood may also affect the accept- ability of such work. In addition, education and childhood residence may affect "tastes. for modern employment, both directly and indirectly, by increasing exposure and susceptibility to such inf luence as mass media advertising. For these reasons, WED and RESC should be positively related to WBM in transitional societies. In traditional settings these relation- ships should be nil due to a presumed scarcity of modern employment opportunities. Early Outcomes Block III of Figure 1.2, corresponding to the second stage of the fer- tility process, contains four early outcomes that have relevance to later events. First, early fertility (EF) refers to the number of children born before respondents reach age 30. Although we anticipate that EF represents relatively uncontrolled reproductive behavior in the countries to be studied empirically, it affects all decisions about later fertility and contraceptive use. Second, survivorship of children during the early phase of the fertility process is also important. ECM denotes child deaths before respondents reach age 30. Finally, the sufficiency of sons and daughters when respondents reach age 30 may influence subsequent reproductive decisions, especially in cultures with strong son preference. SCB is a dummy variable taking the value 1 if the respondent has two or more living boys, and O otherwise. Similarly, SCG takes the value 1 if the respondent has one or more living girls, and is O otherwise. All variables in Blocks I and II are predetermined with respect to both early fertility {EF) and early child mortality (ECM): 60,3 + 61,3RESC + 02,3WED + 83,3AFB + 64 3WBM. (1.4) EF = 00,4 + 61,4RESC + 82,4WED + 63,4AFB + 54,4WBM. Because it is presumed that fertility patterns in the early reproductive years are not responsive to the presence or absence of conscious decision making, major differences in the determinants of EF across societal context are not expected, nor are major differences in the determinants of ECM. m ere is one exception: work experience in the modern sector before marriage (WBM) does not apply in traditional contexts and thus will not affect either ECM or EF in those settings. Childhood residence (RESC), respondent education (WED) , and work experience {WBM) all influence knowledge of hygiene, health practices, breastfeeding patterns, nutrition, exposure to disease, and access to medical care. These variables are expected to have negative relation-
21 ships with ECM (01,3, 82,3, 64,3 < 0) and positive relationships with EF (5l'4' 62 4r 84~4 > 0) ~ except where breastfeeding differentials counterbalance the expected health effects. Expectations about the effect of age at first birth (AFB) are based mainly on exposure. The number of years between onset and age 30 limits births and child deaths t leading to the hypothesized negative relationships between AFB and both EF and ECM (83 3, 03,4 < 0). To the extent that AFB is sensitive to health-related influences over EF and ECM not measured by WED, RESC, and WBM, these expected negative relationships may be dampened somewhat by the negative association between health and AEB. The two sex Buff iciency var tables depend only on EF and ECM: (1.5) SCB = 60,5 + 65,5ECM + 66,5EF (1.6) SCG = 60,6 + 65,6ECM + 66,6EF. Of necessity, the number of children surviving when respondents reach age 30 (EF - ECM) affects the likelihood of having two boys (SCB) or one girl (SCG). Equations (1.5) and (1.6) decompose the effect of survivorship into EF and ECM components, but exclude socioeconomic variables. Sex sufficiency may interact with socioeconomic characteristics in the deter- mination of fertility and contraceptive behavior; however, the probability of a particular birth being male or female does not itself depend on socioeconomic status. Although SES may affect sex preference, which in turn may lead to differential care and nurturing, ECM will mediate its effect on SCB and 5CG. Emus, even as a proxy for sex preference, there is no reason to include SES in equations (1.5) and (1.6~. Endogenous Socioeconomic Variables While education and childhood residence are unambiguously predetermined with respect to onset, early fertility, and later fertility, other socio- economic variables come into play at varying points in the reproductive process as we model it. Work before marriage (WBM? is predetermined with respect to both early and later fertility. Another set of socioeconomic variables is predetermined with respect to later fertility only. Four such variables appear in Block III of Figure 1~2: current residence (RES), respondent postmarital work experience (WSM) f education of current husband (HED), and occupation of current husband (HOCC). The measurement of RES, WSM, and HED parallels that of similarly defined variables in Blocks I and II. Several alternatives can be tried for the measurement of HOCC, including self-employed vs. other, agricultural vs. other r and ordinal scales tapping access to ~modern" work organization. m e treatment of these variables as endogenous is perhaps unconven- tional, but nevertheless justified. Unless the time ordering of a pair of variables is clear and unquestionable, the assignment of causal order- ing is inherently procrustean. The difficulty in assigning endogenous socioeconomic variables to positions in the structural equations is that the time referents of these variables are ambiguous in one way or another. WFS respondents are asked about their current residence and the character-
22 istics of their current husband, where "current" refers to the survey date. There is no way to know how long they have lived in their current residence (not even by matching it with childhood residence), or whether there have been changes in their husband's education and occupation. It can be known whether the current husband is the first one, but not how long he has had his current education and occupation. The time reference for the work since marriage q~est~an is also unclear. The same code is assigned to women who have just started back to work after all their children have reached a certain age and to those who worked only from marriage to the birth of their first child. It might be supposed that if there were continuous work, education, and residence histories available for respondents and their spouses, the correct causal ordering of all the socioeconomic variables could be determined. m is would not, in fact, be the case since this kind of detailed information would generate numerous causal orderings--one for each distinct pattern of transitions. Given an underlying multiplicity of patterns of status transitions, the necessity of working with summary variables, and the ambiguous time referents for those variables, there are two strategies for addressing the question of causal ordering in the model. One is to treat all of the endogenous socioeconomic variables as merely intercorrelated with the variables in Blocks I and II. Although there are instances in which this is appropriate, in the present case we prefer the second strategy, which is to assume a certain amount of causal ordering. This is justified if those causal assignments are consistent or compatible with noteworthy aspects of the broader problem. The placement of RES, HED, and HOCC is problematic with respect to AFB and WBM. The choice of placing these variables in Block TII can be justified on the grounds that it is probably the correct decision for respondents who have been married more than once, and is compatible with the nature of the questions asked, all of which refer to "current" status. mat is, this placement is consistent with an analytic allowance for change. That this placement is erroneous for those whose lives have been quite stable is granted. To the extent that this stability is present, only the direct effects of AFB and WBM, as well as those of RES, HED, and HOCC, have relatively clear interpretations. By the same token, we do not allow ECM and EF to depend on RES, WSM, HED, and HOCC. Again, the dominance of stability in the lives of the respondents, and hence an unambiguous causal ordering, is uncertain. Of course, it is plausible to think of ECM and EF as standing in reciprocally causal relationships with RES, WSM, HED, and HOCC. However, we have not specified the model in this way both because the parameters of such simul- taneous equations would not be identified, and because the particular simultaneities seem relatively uninteresting--even were they estimable. Finally, there seems to be no serious conceptual problem in treating RES, WSM, HED, and HOCC as predetermined with respect to the variables in Block IV, which in the vast majority of instances refer to events taking place after respondent's 30th birthday. The relative clarity of this causal ordering between Blocks III and IV is in fact one of the assets of the processual approach we have adopted. The usual strategy of analyzing cumulative fertility and child mortality would require a simultaneous specification for RES, WSM, HED, and HOCC and the variables of Block IV.
23 m e burden of resolving the identification problem would in this case be considerable; the processual approach involves fewer tenuous assumptions. The equations for the endogenous socioeconomic variables are as follows: (1.7) RES 80,7 + 81,7RESC + 82,7WED + B3,7AFB + IBM. 80,8 + 81,8RESC + B2,~WED + B3 SAFE + B4 IBM. ED Do,g + 81,gRESC + 82,gWED + 63,gAFB + 64 gWBM. (1.10) HOCC = 60,10 + Bl,lORESC + 62,10WED + 63,10AFB + 84,10WBM. m e coefficients of childhood residence, respondent education, and work experience before marriage should be positive in all four equations. Women who grow up in urban areas, who have more than average education, and who work in the modern sector before marriage are likely to find similarly situated husbands, to work after marriage, and to live in an urban area. Except for the inapplicability of modern employment in traditional settings, these expectations hold regardless of context. Age at first birth potentially influences the endogenous socioeco- nomic variables. AFB may capture aspects of respondent background missed by the other socioeconomic variables. In a traditional agrarian setting, childhood residence and education imperfectly discriminate between women of differing socioeconomic status. More revealing indicators might include ownership of land and access to water. Relatively well-off families in these contexts may be able to arrange early marriages for their daughters. AFB may therefore reflect aspects of childhood socio- economic status not measured by RESC or WED. If so, then AFB will be negatively related to the endogenous socioeconomic variables (83,7, 63,9, B3,10 < 0; 83,8 = 0). AFB may also reflect unmeasured aspects of socioeconomic status in transitional societies, although RESC and WED should be improved proxies in these contexts too. The relationships between AFB and the endogenous socioeconomic variables will be positive, however, if our hypothesis that advantaged women delay marriage is correct. The Adjustment Variables Block III of Figure 1.2 contains several adjustment variables that affect later fertility and contraceptive use patterns. Two can be considered together: number of marriages (NMAR) and number of years spent in a union between the 30th birthday and the survey date (DUR). NMAR controls for disruption, whereas DUR accounts for later exposure. The two are modeled as functions of variables in Blocks I and II: {1.11) NMAR = Bo,l1 + B1,11RESC + 62,11wED + 83,11AFB + 64,11WBM- (1.12) DUR = 80,12 + 81,12RESC + 62,12wED + 8 3,12AFB + B4,12WBM.
24 Childhood residence, education, and work before marriage may affect NMAR- and DUR in a number of ways. They reflect norms governing marriage, divorce, and remarriage, as well as differential availability of re- sources. Wealthy families may be better able to afford dissolution and remarriage. On the other hand, poor families are at greater risk of dissolution due to death. mese relationships depend upon the particular society studied. Hypotheses about the effects of age at first birth are also society-specific, however, there is no reason to suppose that for either the NMAR or the DUR equation, coefficient variability between societies is related systematically to the traditional/transitional continuum. A third adjustment variable is self-reported fecundability (FEC), a dummy variable where the value 1 indicates no known problems. Perceptions of fecundability (at the time of the survey) are undoubtedly connected with proven fertility and survivorship. This variable is considered contemporaneous with early outcomes and assumed predetermined with respect to later outcomes. It is likely to depend on luck, which is random, and health, which is tapped by variables in Blocks ~ and II: (1.13) FEC = 60,13 + 81,13RESC + 82,13WED + 63, 13AFB + 64,13WBM. Age at first birth is also predetermined with respect to FEC, even though its effect can be positive or negative. On the one hand, late onset may indicate subfecundity, producing a negative relationship between AFB and FEC (03,13 < 0~. On the other hand, because late onset leaves little time for women to judge their fecundability, their self-report may be optimistic {83,13 > 0~. As with the NMAR and DUR equations, intersociety coefficient variation for the FEC equation should not be related to the traditional/transitional continuum. Later Child Mortality Block IV of Figure 1.2 contains three endogenous variables, one of which is later child mortality {LCM). This refers to child deaths occurring between the 30th birthday and the survey date, although the children could have been born before respondents reached age 30. All of the variables in Blocks I through III can determine LCM: (1.14) LCM = 60,14 + 01,14RESC + 02,14WED + 63, 14AFB + 64,14WSM 65914ECM ~ 56,14EF + 8 7,14SCB + B8,14SCG + 69,14=S + 610,l4WSM + 611,14=D + 612,14H°CC + ~13 :4NMAR + 514,l4DUR + 61S,14FEC Major differences across context in the estimated effect parameters seem unlikely (although average number of child deaths will differ). Justifications for the expected effects of the predetermined vari- ables in equation (1.14) fall into two broad categories. The first
25 concerns the size and age-sex composition of the early family. Older children enjoy lower mortality risks than younger ones. Controlling for EF and ECM, an early age at first birth implies older children and lower later child mortality (83~14 > 0)e There should be a positive relationship between EF and LCM (06r14 > 0) because larger families will have more children at risk and place a greater strain on family resources than smaller families, other things being equal. It is also possible that, apart from family size, sex composition influences the distribution of family resources among children in some cultures. We include SCB and SCO so that this hypothesis can be examined; their hypothesized effects are culture specifics m e rationale for the second category of expected effects involves health conditions. Residence, education, work experience, and occupation influence exposure to disease, health-related behavior, and access to professional care. The health implications of the socioeconomic vari- ables will in all likelihood influence later child mortality (01,14, 82,14, 64,14, Bg,l4, 010,14, 611,14, B12~14 < 0). Women believing that they are no longer fecund may invest more time and care in children than fecund women, since the latter can replace lost children (615,14 > 0) For genetic and family-specific reasons, there should be a positive relationship between early and later childhood mortality (85,14 > 0). Number of marriages and time spent in a union may affect resources for child care and the equity of distribution across children, suggesting (613,14 > 0; 614,14 < 0~. It is also true, however, that OUR, measured as time in union since age 30, will be associated with the number of children born after age 30, and will therefore reflect the number of infants exposed to the risk of dying (814,14 > 0). m us, the effect of marital duration (614,14) may be either positive or negative. We assume it to be negative for the purposes of the present discussion, but have determined that the sign of this coefficient has only minor consequences for the implications of the model (derived in Section 1.4~. Contraceptive Use Patterns Other things being equal, women who use contraception will end up with smaller families than those who do not. It is probably fortunate that this proposition is generally thought to be self-evident, since it would be difficult to demonstrate it empirically using the WFS and other similar survey data. m e problem is that the proposition contains an implicit temporal ordering not typically measured in sample surveys of fertility behavior. If it is unclear when contraception began and how consistently it was used, it is possible to obtain a positive correlation between current or ever use of contraception and fertility. This will occur in transitional settings because the decision to use contraception is based in part on the number of living children at the time. Thus, not only is the association in the wrong direction, but also the flow of causality is in both directions (between fertility and contraception) rather than unidirectional (from contraception to fertility). What, then, is the point of including contraceptive use as typically (incompletely) recorded in sample surveys such as the WFS? One answer,
26 the answer that motivates our strategy, is that one must avoid the attempt to show that contraception works and instead concentrate on the forces that lead to differentials in contraceptive uses. A positive correlation between, say, current use of contraception and fertility is irrelevant. We know why it occurs, and we know also that there are efficient methods of contraception, which is to say that it works. The pressing task is to increase our understanding of the processes that lead to use and to estimate the relative magnitude of their effects. In this connection, our partitioning of children ever born into early and later fertility again proves useful. Because contraception in the early years of childbearing is relatively uncommon in transitional societies, becoming more common In the later years, it is plausible to treat (incompletely} measured contraceptive use as endogenous with respect to early fertility. As noted above, this makes the model much more precise and substantively meaningful than if the basis were children ever born (the commonly used measure). The essential point is that contraceptive use as typically measured (e.g., current or ever use} is simultaneous not only with children ever born, but also with later fertility. In our view, the requirements for identifying this simul- taneity cannot be satisfied reasonably,--given the kind of information available in the WFS and other similar surveys. -The decomposition of children ever born resolves this problem. Similarly, there is a simul- taneity between child mortality and contraceptive use as typically measured. me division of child mortality into its early and later components also resolves this simultaneity. It is meaningful to treat early child mortality as predetermined with respect to contraceptive use in the data to be used for our empirical studies. There is no one best way to operationalize contraceptive use in the WFS data. Indeed, LOU is more difficult to measure than any other vari- able in the model, because of both the limitations of the WAS question- naire and the inherent complexity of contraceptive behavior. The WFS data on contraceptive use are rather less than a complete contraceptive history. At most they include use of efficient and inefficient methods currently, in the open birth interval, in the last closed birth interval, and ever. Our goal is to use this information somehow to tap the "modernity" of contraceptive behavior. To this end, we are experimenting with an ordinal scale for contraceptive use that attempts to distinguish long periods of effective use from short or ineffective periods of use, based on lengths of the open and closed intervals among current users. mere are several challenges to the successful use of this or other contraceptive use scales based on WFS data. First is the distinction between efficient and inefficient methods. The latter are apparently being used with a high degree of success in the Philippines (Williamson, 1982), and perhaps elsewhere. For this reason, it may prove worthwhile to consider use of such methods (e.g., rhythm) in addition to efficient methods such as the pill and IUD. Second is sterilization, which poses a different kind of problem. We propose grouping contraceptive steriliza- tion with efficient contraceptive methods, and reserve for empirical study the problem of what to do with the very small number of noncontraceptive sterilizations. Our treatment of this problem may vary across countries.
27 There remains the question of what to do about abortion in the context of a contraceptive use scale. Although some WFS countries included an induced abortion question, many did not. It appears that abortion frequencies are not low in countries where the abortion question was not asked (Tietze, 1981~. Even where the question was included, the data are likely to severely underestimate the frequency of abortion (Srikantan, 19821. An ~ndividual's past use of abortion is not highly predictive of future use of abortion. Consequently, Current or potential use of abortion cannot be ascertained. For all of these reasons we plan to exclude abortion from individual-level measures of contraceptive use, although we may be able to incorporate some of its effects through the inclusion of contextual variables. The distinction between traditional and transitional settings is based on the concept of discretionary fertility behavior, which is implemented primarily through contraception. m us expectations about the effects of the predetermined variables on LCU apply only to transitional settings. m e following equation summarizes relationships involving the contraceptive use patterns depicted in Figure 1.2: (1.15) LCM = 80,15 ~ 5l,15RESC + 62,1sWED + 63, 15AFB + 64, 15WBM + ss,1~CM + IF + c7,1~CB + 68,15SCG + 69,15RES + 610,15WSM + 611,15HED + 312,15H°CC + 613 15NMAR + 614,lsDUR + B15,15FEC ~ 616,15(AFB) 617,l5EF WED + 818,lsEF*HoCC + 819,15SCB*WED + 820, 1sSCB*HOCC. As can be seen, the prediction of LOU is considerably more complicated than that of other endogenous variables considered thus far. Onset, early fertility and mortality outcomes, socioeconomic characteristics, and the adjustment variables may all influence contraceptive use patterns. In addition, two sets of interaction terms appear in equation (1.15~. We hypothesize that onset plays a dual role in the determination of LOU. On the one hand, a young AFB may signal compliance with traditional norms and ideals. To the extent that it does, it will imply a similar compliance in contraceptive practice. On the other hand, late AFB could also be linked with low contraceptive usage because it might indicate subfecundity, lessening the need to contracept. Although FEC appears in equation (1.~5) , it may not capture this possibility completely. In addition, women who are fecund may be less inclined to control their fertility after a late AFB. Fey may be aware that the probability of secondary sterility will increase as they age, and may wish to insure against this possibility. mese arguments suggest a curvilinear relationship between AFB and LOU (53,15 > 0; 816,15 < 0) Contraceptive use patterns should be sensitive to boy and girl sufficiency (SOB and SCG) and the size of the early family. Early fertility and mortality outcomes, judged against ~ family-size target,
28 will affect the subsequent use of contraception. EF should be positively related to LCU: respondents who are close to their targets when they reach 30 are likely to adopt effective birth control methods thereafter, and will do so earlier than their less prolific counterparts (06,15 > 0) Of course, early family size depends on child deaths as well as births. Early experience with child mortality (ECM) should diminish the proba- bility of subsequent contraceptive use (B5~5 < 0~. LCU may respond to sex composition as well (as reflected in the "sufficiency variables). Where son preference is strong, repondents may not even consider family- size limitation until they have achieved a ~sufficiency" of boys, here operationalized as two living sons. Respondents may also have a strong preference for at least one daughter (Arnold et al., 1975~. If so, the presence of two living sons (SCB) and one living daughter (SCG) should increase the likelihood of subsequent contraceptive use {87,15, 68 15 > 0), although the magnitude of this impact will undoubtedly vary from one country to another. If our hypothesis about the effect of EF on LCU is correct, the strength of the relationship between the two will depend on family-size targets. These, in turn, may depend on socioeconomic characteristics. mere is thus a possibile interaction between EF and socioeconomic variables; in this case, respondent's education (WED) and husband's occupation (HOCC) are selected as likely candidates. One way to inter- pret these interactions is in terms of innovative behavior. Innovators will be even more likely to use contraception than would be expected based on their early family size and socioeconomic characteristics (017,15' 818,15 > 0, assuming innovation among relatively advantaged groups). m e magnitudes of the interaction effects may reflect how much diffusion of new behavior patterns to less innovative groups has occurred. There is also a possibility of culture-specific interactions between socioeconomic characteristics and SCB. Socioeconomic characteristics should also be related to knowledge of modern birth control methods, willingness to use them, and efficiency of use. Current residence (RES) has implications for contraceptive knowledge, attitudes, and practice. In particular, mass media campaigns designed to promote the use of birth control are often concentrated in urban areas. Urban residents are also likely to have access to media channels (e.g., radios). Specific information may be disseminated not only via the mass media, but also by classroom instruction or even special programs in the workplace (e.g., sterilization campaigns in India). Education, respondent's work experience, and occupation can influence LCU in this way. Moreover, socioeconomic characteristics define reference groups and role models: educated urban dwellers in modern occupations are more likely than their traditional counterparts to have friends, neighbors, and peers who practice effective contraception. m ese expectations are summarized in the hypothesis that contraceptive use patterns vary positively with socioeconomic status (01,15, 62,15, 84,15' 69,15' 610,15' 611,15, 612,15 > 0) We include the adjustment variables in equation (1.15~. Women who think they are subfecund have no reason to adopt contraception (B15,l5 > 0~. Respondents still in their first marriage or union at the time of inter- view are more likely to be losing birth control than those in higher-ord--
29 unions ( ~ 13,15 ~ 0 ) . Quite of ten , children symbolize the existent e of a marriage and the joining of two families across future generations. Family-size targets may apply within marriages, with women who partici- pate in several having less incentive to contracept than otherwise. Moreover, some exposure time is lost as one marriage dissolves and another comes into being, further weakening the incentive to contracept (014,15 > 0~. In addition, since contraceptive practice has increased over time in many WFS countries (Johnson-Aceadi and Weinberger, 1982), respondents with low recent exposure are less likely to be users of efficient means of contraception. The inclusion of DUR and NMAR helps adjust for this, so that the effect of other variables is less distorted. Later Fertility Later fertility (LF) refers to children born between the 30th birthday and the survey date. Variability in LF will occur for women at least 35 years old when interviewed, and will increase with respondent's age. For this reason, Block IV requires older cohorts of women. m e following equation summarizes the effects of predetermined variables on LF, depicted in Figure 1.2: ( 1 . 16 ~ LF ~ 0, 16 + ~ 1, 16RESC + ~ 2, 16WED + ~ 3, 16AFB + ~ 4, 16WBM + 65,1~CM + 66,1~F + 67,16=B + 68,16=G + 69,16~S + ~ 10, 16WSM + ~ 11, 16HED + ~ 12, 1tHOCC + 613, 16NMAR 14, 16DUR + ~ 15, 16FEC + ~ 16, 16 (AFB) 2 + ~ 17 16EF*WE D + ~ 18, 16E:F*HOCC + ~ 19, 16SCB*WED + ~ 20, 16SCB*HoCC ~ Equation {1.16) includes the same predictors as equation (1.15~. As will be seen, however, contextual differences in our hypotheses about constitu- ent relationships are much more pronounced for LF than for LOU. More individual discretion is possible with LF than with any earlier components of the fert' 1` ty process, In transitional settings, LF should be a func- tion of family-size targets. Factors affecting the attainment of these targets should also have an influence. In traditional settings, fam~ly- size targets are less germane. Indeed, there is some question whether well-defined numerical targets exist at all (Caldwell, 1977~. To the extent that socioeconomic characteristics and early outcomes affect LF in traditional contexts, they should do so via consequences for exposure, fecundability, and health. Our hypotheses about the signs of almost all of the coefficients in equation (1.16) depend on societal context. as discussed more fully below. Tr ens i t iona l Borate - at ,; In transitional populations, there should be an inverse relationship between socioeconomic status and LF, other things being equal. We view
30 socioeconomic characteristics as critical determinants of implicit family-size targets. Respondents who grew up in urban areas, live in urban areas at the time of the survey, received some schooling, worked in the modern sector before and after marriage, and are married to educated and well-off husbands are likely to want small families and have low later fertility (~31~16' 632~16, 64~16, 639~16r 6310~36, Allele, `312~16 ~ 0~. This is due to exposure to small-family ideals and norms, and perhaps to a decline in the influence of significant others who hold traditional norms and ideals. Given a specific family-size target, women having a large family by their 30th birthday are closer to their target than are those with smaller families. We therefore hypothesize an inverse relationship between early and later fertility in transitional contexts (06,16 < 0~. High rates of child survival imply closer proximity to a target (05,16 > 01. Hey may also reassure parents about future mortality prospects, thereby reducing the need for "insurance" births. Family-size targets can involve desired minimum numbers of boys and girls, with the boy and girl sufficiency of the early family expected to have the greatest impact in cultures with a strong son preference (07,16' 68,16 < 0~. Since the strength of son preference can vary across socioeconomic groups within cultures, we include interactions between SCB and two socioeconomic characterist~cs-- respondent's education {WED) and husband's occupation tHOCC). We also include interactions between these two socioeconomic variables and EF. mese terms denote the possibility that the coincidence of small-family norms and close proximity to a target produces a stronger fertility response than that measured by an additive model. AS was the case in the LOU equation, these interactions can also be interpreted in terms of innovative behavior. There should be a curvilinear relationship between AFB and LF, such that early and late onset correspond to more children after age 30 {03,16 < 0; 516,16 > 0~. The rationale for this specification is the same here as for LCU. It is self-evident that women who are fecund have higher fertility later in life than their less fecund counterparts, other things being equal (015,16 ~ 0~. The adjustment variables control for exposure (~4,16 > 0) and the fertility-enhancing effect of several unions (313,16 ~ 0) Traditional Contexts The expected effects of the predetermined variables are quite different in traditional contexts. Indeed, hypotheses about the impact of self- reported fecundability and adjustment variables are the only ones invariant across context. We assume that in traditional settings, early and later fertility are equally unregulated. Consequently, respondents with high early fertility will probably end up with more children later in life than their less prolific counterparts (06,16 > 0~. Early child mortality is not expected to exert much influence because any response would imply fertility control (05,16 = 0~. For the same reason, boy and girl sufficiency should matter little (07,16' 5~,16 = 0~. The absence of relationships between early outcomes and LF removes the need for inter- action effects (817,16' 018,16, 619,16, 620,16 = 0)
31 mere should be a weak positive relationship between socioeconomic status and later fertility in traditional settings. This hypothesis derives from the positive health effects implied by relatively advantaged social positions. Husband characteristics may prove vital in measuring family well-being (B11 16, B12,16 > 0), especially nutritional levels and overall health. Healthy women are more fecund. AFB may also reflect social standing (63,16 < 0), as argued earlier. However, there is no basis for positing a curvilinear relationship between AFB and LF in traditional societies; hence in these settings, we would set 816,16 = 0. Because of their correlations with health and fecundity, respondent char- acteristics should also affect LF (~l,16, 82,16, 89,16, 610,16 > 0). Each individual relationship should be weak both because there are so many, and because they imperfectly approximate the underlying, unmeasured variables. 1 .4 REDUCED AND SE==DUCED FOR IMPLICATIONS OF THE MODEL His section discusses some implications of the model, first explaining why its complexity needs to be reduced. Next, certain implications of the model for three familiar variables--children ever born, contraceptive use, and infant and child mortality--are traced. This leads to develop- ment of implied and actual semireduced form equations for these variables, followed by a discussion of inferences that can be drawn about the coefficient signs in these equations. The Need for Reduction As shown above, the threefold distinction between onset, early fertility, and later fertility permits development of an intricate and comprehensive micro model embodying a number of untested hypotheses. This micro model is not the total model, which would also incorporate macro variables to explain coefficient variability across settings. We intend to continue the theoretical development of the model begun in this paper elsewhere. Our present focus is on testing the implications of the micro model. As a starting point, Table 1.2 presented the predicted signs of the structural coefficients in traditional and transitional settings implied by the micro model. How is the model to be tested? Its estimation for a single country would permit one kind of test. In particular, where we have hypothesized that a coefficient should have a positive sign in all settings, the hypothesis could be rejected empirically if the sign were found to be in the opposite direction. However, this would be a weak test. Among other things, it applies only to those coefficients for which a single sign was hypothesized. How can hypotheses be tested which relate to coefficients for which sign variation across settings is possible? There will be variation not only in the sign, but also in the magnitudes of the coefficients. His variability is theoretically admissible and is implicitly assumed to exist in the above discussion, since for a coefficient to Travels from negative to positive, it must take on a
